FEED FORWARD BACK PROPAGATION NEURAL NETWORK METHOD FOR ARABIC VOWEL RECOGNITION BASED ON WAVELET LINEAR PREDICTION CODING

A novel vowel feature extraction method via hybrid wavelet and linear prediction coding (LPC) is presented here. The proposed Arabic vowels recognition system is composed of very promising techniques; wavelet transform (WT) with linear prediction coding (LPC) for feature extraction and feed forward backpropagation neural network (FFBPNN) for classification. Trying to enhance the recognition process and for comparison purposes, three techniques of WT were applied for the feature extraction stage: Wavelet packet transform (WPT) with LPC, discrete wavelet transform (DWT) with LPC, and WP with entropy (WPE). Moreover, different levels of WT were used in order to enhance the efficiency of the proposed method. Level 2 until level 7 were studied. A MATLAB program was utilised to build the model of the proposed work. The performance of 82.47% recognition rate was established. The mentioned above methods were investigated for comparison. The best recognition rate selection obtained was for DWT.